We need to try check table parameters during registration endpoint call, instead of currently asynchronously check during state sync loop.

This way the issues like non-existent users on the source cluster would be obvious and easy to debug.